Welcome to the [AI Daily] column! This is your guide to exploring the world of artificial intelligence every day. Every day, we bring you hot topics in the AI field, focus on developers, and help you gain insight into technology trends and learn about innovative AI product applications.
Fresh AI products Click to learn more: https://app.aibase.com/zh
1. Volcano Engine launches the Doubao real-time audio and video call model SeedRealtime 1.0 API
Volcano Engine has officially launched the API service for the Doubao real-time audio and video call model SeedRealtime 1.0. It adopts end-to-end unified modeling, breaking the bottleneck of traditional ASR+VLM+TTS cascading, and enabling real-time multimodal interaction that can watch, listen, and speak simultaneously, providing more natural and smooth response capabilities for scenarios such as intelligent assistants and real-time companionship.

[AiBase Highlights:]
🎯 End-to-end unified modeling eliminates cascading latency and information loss
⚡ Enables watching, listening, and speaking simultaneously, improving the naturalness of multimodal interaction
🔗 Open API service empowers intelligent assistants and real-time companionship scenarios
2. Tencent sharply cuts investment in Yuanbao, WeChat AI assistant "Xiaowei" reaches 30 million daily active users
Tencent's consumer-facing AI strategy has clearly shifted toward pragmatism: Yuanbao has been significantly cut back due to failing to meet expectations, with focus returning to model capabilities; WeChat Xiaowei is accelerating expanded testing based on the ecosystem, with daily active users already reaching 30 million and lower costs. Overall, the approach is shifting from burning money for scale to prioritizing efficiency and foundation.
[AiBase Highlights:]
📉 Yuanbao failed to meet expectations, investment was sharply reduced, and focus shifted to model capabilities
💬 WeChat Xiaowei accelerates expanded testing, daily active users have reached 30 million
💰 After launch, Xiaowei's quarterly cost is lower than Yuanbao's earlier level, with greater emphasis on efficiency
3. Google releases Gemini 4 Neon, speed soars 3x and it also supports a million-token context
Gemini 4 Neon is positioned for complex reasoning and multi-step task automation, with outstanding million-token context and multi-agent scheduling capabilities. Its coding agent index surpasses Claude Opus 5.5. Early evaluations show task quality is on par or better, with 3x faster speed and lower cost. However, the benchmark tests have limiting conditions, and actual business performance still needs to be observed.

[AiBase Highlights:]
🧠 Million-token context window supports comprehensive processing of multimodal data
💻 Coding agent index surpasses Claude Opus 5.5, ranking first
⚡ Speed reaches 3x that of Carbon, with per-task cost reduced by 32%
4. Anthropic plans IPO pricing at $83 per share, fully diluted valuation reaches $1.5 trillion
Anthropic priced above its upper limit, and the $1.5 trillion valuation highlights the scarcity premium of leading generative AI companies; Amazon, Google, and chip giants are deeply tied in, strengthening ecosystem synergies. Despite sector volatility, the high pricing still reflects institutional optimism about commercialization prospects and provides a valuation anchor for subsequent IPOs.
[AiBase Highlights:]
💰 Issue price of $83 exceeds expectations, fully diluted valuation reaches $1.5 trillion.
🤝 Amazon and Google hold nearly 15% and 14% stakes respectively, and tech giants subscribed to $26 billion in shares.
📊 Listing performance will establish a valuation reference for unlisted leading AI companies such as OpenAI.
5. Moonshot Kimi executive appears in a maid outfit at the 2026 Yunqiao Conference, seemingly cosplaying DeepSeek's "Big Fat Fish"
Although competition in the AI circle is fierce, developers using anime-style personas to create lighthearted topics reflects the cultural diversity and inclusiveness of the tech community. Tang Feihu's hardcore competition and engineering background show that personal expression and professional ability can coexist, and it also carries a certain degree of brand marketing significance.

[AiBase Highlights:]
🤖 Kimi engineer Tang Feihu appeared in a maid outfit at the Yunqiao Conference
🐳 The look seems to imitate DeepSeek's anthropomorphic image "Big Fat Fish"
🏆 Has won an ACM gold medal and the Microsoft Beauty of Programming championship, with hardcore technical strength
6. AI assistant competition shifts to work context management: Tencent WorkBuddy reconstructs task organization with groups and multi-folder associations
Tencent WorkBuddy changes the task list into grouped work directories, supports drag-and-drop and pin-to-top sorting, and allows a single group to associate up to 16 local folders and designate a primary folder. It unifies task organization and local material context at the group level, reducing the cost of cross-folder retrieval and marking the evolution of AI assistants toward sustainable workspaces.
[AiBase Highlights:]
🗂️ The original "Tasks" is renamed "Recent," "Spaces" is upgraded to "Groups," and the sidebar is integrated by work items.
📎 A group can associate up to 16 local folders and designate a primary folder for priority reference.
📌 Groups and tasks support pinning and manual sorting, making high-frequency multitasking more convenient.
7. OpenAI's first AI hardware reportedly to begin shipping by the end of the year: Chinese supply chain manufacturers such as Luxshare Precision are deeply involved
Supply chain information shows that OpenAI's first on-device AI hardware has entered mass production preparation. Luxshare Precision is solely responsible for the complete device and core components, reflecting the key position of Chinese manufacturing in the AI hardware chain. If it adopts screenless voice interaction, it means the AI entry point extends from software to sensory terminals, but the final specifications still await official confirmation.
[AiBase Highlights:]
🚀 OpenAI's first AI on-device hardware is expected to begin shipping by the end of the year
🏭 Luxshare Precision is solely responsible for final assembly and supplies core components
🍩 The product is suspected to adopt a screenless "donut" design, focusing on voice interaction
8. DeepSeek adjusts API pricing strategy: server-side search and web scraping officially exempt from separate billing
This strategy adjustment reflects the pricing approach of large model platforms shifting from value-added services to integration of foundational capabilities. It effectively reduces the multi-round retrieval costs of internet-connected agent applications and has a positive incentive effect on the developer ecosystem.
[AiBase Highlights:]
🔍 Server-side search and web scraping are no longer charged separately
🪙 Billing shifts entirely to model token usage only
🚀 Reduces the cost of internet-connected AI applications and promotes the popularization of agents





